5 basics of big data

June 13, 2012

At the recent HIMSS Virtual Conference and Expo, Chris Gough, solutions architect at Intel Healthcare Information Technology and Alan Stein, MD, vice president of healthcare technology Autonomy, an HP company, presented a webinar titled, "Big Data and Analytics in Healthcare."

Gough and Stein outlined five basics of big data.

1. The main problem is the fragmentation of data. The separation of data among labs, hospital systems, and even clinical components, like financial IT and EHRs, serves as the main issue with leveraging the data, said Stein. "All of these are separate repositories for information," he said. "Their single use in nature is to provide clinical care or provide scheduling information or operational information, and this is a problem if we want systems to talk to each other." Sometimes, he added, an organization can also end up with redundant information due to a legacy system. "So we also have this normalization problem," he said. "And this is where we want to go: we want to improve quality of care and lower costs…we need a shift from best practices to a culture of best practices – if we have them available – but also best experiences and using data from various components of health IT to improve care and lower costs in a holistic way."

2. Big data is all about real or near-real time. Traditional analytics, said Gough, use ETL processes that upload data nightly or weekly to a data warehouse. Processing takes place in the warehouse, yet, the trend of big data is moving toward real or near-real time. "It's not waiting for batch processes but is driving value from data more immediately," he said. "In healthcare, it's clinical decision support, so at the point of care, being able to understand data to make a decision." With traditional analytics, Gough said, reporting focuses on the past, but with big data, "it's more predictive, and it looks forward to what may happen in the future."

[See also: Big data: opportunity and challenge.]

3. Processing is moving to the data. Another trend Gough pointed out is the processing coming to the data, instead of the other way around. "So traditionally, you move data out of a production database to a warehouse, and you pull from different repositories." At the rate data is increasing in healthcare though, he said, whether it's from medical imaging, EHRs, etc., moving this data around is becoming more of a challenge. "So the trend we're seeing is moving processing to the data," said Gough. "That's a large job, that's split up into a number of parts and split across a system. The infrastructure knows where the data resides, and processing happens as close to it as possible to improve performance."

Showing 2 Comments

Matthieu Blit say: Data quality

October 30, 2012 | 07:06 AM GMT

Hello.
Really interesting post, as it points out original aspects of the Big data management topic.
I would also add the data quality challenge, the ones entered by the physicians, as it seems to be quite an issue (http://thehealthcareblog.com/blog/2012/10/04/the-data-entry-paradox/#mor...).
And mentioning the small practices is also important, as they are usually forgotten when it comes down to information management (in general) and big data management (in particular).

Matt Hargus say: Mostly spot on

June 21, 2012 | 12:26 PM GMT

Overall, I would agree with the points made in the article. Data islands are a key roadblock to comprehensive analysis, and horizontal scaling has been a growing trend over the past decade.

I feel the benefits of real-time analytics are focused on a few key areas in the industry, and that instead applying our efforts to analyzing historical trends in an attempt to project a vision of the future results will have a more wide reaching impact. I'm an advocate for real-time processing when practical and useful, but not as a global initiative.

Additionally, while SaaS-based systems tend to gain a footing initially in smaller businesses due to their low implementation costs, to limit them to that market is short sighted. SaaS can have as great, if not greater, an impact for large organizations, often allowing for companies to shift resources from maintaining internal legacy applications back to the core operations. Obviously this assumes that the folks building the SaaS applications have done their jobs well!

Again, I feel you did a fine job of summing up the often hazy world of Big Data as it relates to healthcare. I would love to see future articles focusing in depth in the same area.